Particle-Based Message Compression for Cooperative Localization

• Published in: 2016 IEEE 84th Vehicular Technology Conference (VTC-Fall) pp. 1 - 5
• Main Authors: Mendrzik, Rico, Lewandowsky, Jan, Bauch, Gerhard
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2016
• Subjects: Bayes methods; Belief propagation; Distortion; Mutual information; Noise measurement; Random variables; Wireless networks
• DOI: 10.1109/VTCFall.2016.7881147

We propose a novel message compression scheme for distributed nonparametric belief propagation in the context of cooperative localization. The messages to be exchanged in nonparametric belief propagation are given as particle representations. Communication constraints on the links make the transmission of entire particle representations prohibitive and require compressed (approximated) messages. We tackle the problem of compressing messages in an information-theoretic manner. For that reason, we define a relevant random variable and consider the mutual information shared with it for particle selection. We show that the shared mutual information depends on the choice of the particles. We propose a particle-based message compression algorithm that flexibly trades localization accuracy for communications, i.e. we incorporate a scalar parameter to adjust the degree of compression. We show that the proposed message compression method outperforms its parametrized counterpart in terms of localization accuracy for a well-chosen degree of compression, but comes at the expense of somewhat increased communications.